"We've begun a very exciting cooperation," says Kikki Kleiven, Director of the Bjerknes Centre for Climate Research.
She is visiting the Norwegian Ministry of Education and Research, with the US Department of Energy, to evolve a partnership on machine learning and AI, Artificial Intelligence.
At a meeting in Oslo this week researchers and Bjerknes partners at the Nansen Centre, NORCE and University of Bergen are present to present their future project plans in climate modelling, using machine learning and algorithm training.
Projects are being built to improve climate prediction with algorithm trained on understanding patterns in weather and wind systems, or to understand how sea ice moves and breaks. Machine learning can also be used to develop socalled super models, a combination of several models, where you can correct for known model errors. The result means methods that are closer to observations than each model separately.
An Earth defined in equations
A climate model can be viewed as a laboratory of the earth and its climate system, defined in equations in a computer program. Climate models are one of the main tools of climate research, but they do have their limitations.
Limitations like the understanding of processes in the climate system, how the processes are represented in the climate model, couplings and feedback between different processes. While these limitations are about the basic research and scientists' understanding of the climate system, there are some limitations where new methods can shine: machine learning.
In carbon observation machine learning is used with observations to estimate the oceans' annual carbon uptake.
Handling large amounts of data
Through runs of climate models enormous amounts of data is being produced. In addition, international model comparisons make sure we have model results with better probability. Large model comparisons like CMIP (Coupled Model Intercomparison Project) form the foundations of the main reports from IPCC.
"With data amounts like this, you'd have to use large super computers. These are costly model runs, and the data are close to unmanageable in size. A partnership like this can take climate modelling many steps forward," says Kikki Kleiven.
